Solving a big-data problem with GPU: the network traffic analysis

Abstract

The number of devices connected to the Internet has increased significantly and will grow exponentially in the near future, it is due to the lower costs. It is expected that next years, data traffic via Internet increases up to values around zettabyte. As a consequence of this increase, it can be observed that the data traffic is growing faster than the capacity of their processing. In recent years, the identification of Internet traffic generated by different applications has become one of the major challenges for telecommunications networks. This characterization is based on understanding the composition and dynamics of Internet traffic to improve network performance. To analyse a huge amount of data generated by networks traffic in real time requires more power and capacity computing. A good option is to apply High Performance Computing techniques in this problem, especifically use Graphics Processing Unit (GPU). Its main characteristics are high computational power, constant development and low cost, besides provides a kit of programming called CUDA. It offers a GPUCPU interface, thread synchronization, data types, among others. In this paper we present the causes of increasing data volumes circulating on the network, data analysis and monitoring current techniques, and the feasibility of combining data mining techniques with GPU to solve this problem and speed up turnaround times.